rose.blog

Back from choking some folks out. Gots ta represent.

OK in this second installment of Geordie’s Easter Weekend Vent-a-thon, I’d like to offer some food for thought related to Scott Aaronson’s latest blog post. Specifically I’d like to take him to task on his rationalization of why he is so crusty, the most recent examples of which are:

1. I asked Geordie Rose (the founder of D-Wave and sometime commenter on this blog) twice for actual information about the speedup D-Wave claimed to be seeing over classical algorithms, and he never provided any.

I don’t remember actually ever receiving this request from Scott, so I apologize for not being timely in my response. For the record, here is what I have said, in some of the posts here, several interviews, and the demo itself: (A) Orion is roughly 100x slower than state of the art algorithms running on a $1000 PC for solution of small Maximum Independent Set problems, and (B) the only way scaling can be extracted is empirically, and we can’t build big enough systems (yet) to answer scaling questions.

2. Instead of answering my and others’ technical questions, D-Wave decided to attack academic science itself. Here’s what Ed Martin, the CEO of D-Wave, told an interviewer from ITworld: “Businesses aren’t too fascinated about the details of quantum mechanics, but academics have their own axes to grind. I can assure you that our VCs look at us a lot closer than the government looks at the academics who win research grants.” The track record of companies that engage in this sort of behavior is not good.

I can’t see how you can argue that we haven’t tried to answer technical questions. I go out of my way to answer any questions that are asked. If you have any let me know, I’ll answer them if I can.

Casting Herb’s (not Ed’s-there’s something ironic there) statements as attacks on academic science is ridiculous. The above statement has been taken WAY out of context. The original question was how our investors do due diligence on the company without any peer reviewed publications, which is a fair question. Herb’s answer–which is a true statement, whether or not you like it–is that financing a company like this puts us under levels of scrutiny far beyond the norm, either for a start-up company or a big research grant. Herb’s statement isn’t an attack on academic science. It’s that the bar we have to get over to raise cash is higher than it is for other start-ups or academic groups.

I don’t know what you mean by that ominous last statement. You mean like Intel, Microsoft, IBM, GE?

3. I read article after article claiming that quantum computers can solve NP-complete problems in polynomial time. I reflected on the fact that a single one of these articles will probably reach more people than every quantum complexity paper ever written.

This is probably true, although only 50 people in the world understand what you just said. And note that I have never said that, and in fact go out of my way to state a point of view that is very similar to Scott’s.

4. It became clear, both from published interviews and from talking to Jason, that D-Wave was doing essentially nothing to correct journalists’ misconceptions about what sorts of problems are believed to be efficiently solvable by quantum computers.

While I personally find questions about efficiently solvable problems fascinating, these issues are remote from what we are actually doing here.

This is worth emphasizing, because I thought it was obvious, but it turns out alot of people don’t get this. Most of the poorly thought out comments related to what we’re trying to do have come from theoretical computer scientists, who assume that the things they hold dear are likewise treasured by everyone else. Because they worship efficiency, they have assumed that’s the objective of our projects, when I have repeatedly said it’s not.

Here is what I care about: (A) getting to the same level of accuracy as state of the art heuristics in less time and/or (B) given an allotment of time, produce a more accurate solution than state of the art heuristics, for high-value discrete optimization problems that people actually care about.

When people ask what the systems we’re building are for, I tell them that they are for solving discrete optimization problems (which they are). For someone for whom complexity theory is the fulcrum of the universe, this might be interpreted as a statement that we plan to exactly solve all discrete optimization problems, efficiently. For most people who have to deal with these types of problems in real life, the questions arising have a slightly different flavor: “Can you beat our simulated annealing approach?”, which seems like a lot more intelligent question.

…can be found here. A version of this article is to appear in the April 8th issue of the New York Times, which is absolutely great.

Articles about contentious issues, with people seeming to take opposite sides of something that sounds important, get a lot more readers than ones where everyone agrees about something. In that spirit I would like to make a few comments directed at the folks who are being portrayed as being somehow opposed to our effort or its objectives (like Scott Aaronson, Umesh Vazirani, etc.).

Let me start be quoting a passage from the TR article.

Of Geordie Rose’s claims to having built the first practical quantum computer, Aaronson wrote in an e-mail, “Whatever else D-Wave might or might not have done, this can be instantly rejected as hype. If by ‘practical’ he means able to solve practical problems faster than existing classical computers, then this is clearly false. If he means able to solve tiny demonstration problems, he’s been beaten by loads of people. So I can’t think of any interpretation under which he’s telling the truth.”

It’s my view that we have built the first practical quantum computer, so let me present an argument in support. Let’s say Scott, as an intelligent non-expert (in experimental physics), wanted to run a program on a real quantum computer, and for the sake of specificity let’s compare a liquid state NMR QC vs. Orion, without any outside help except for any documentation he could find. The problems associated with converting an algorithm to an actual NMR QC pulse sequence that does what you want are Significant. In fact we enlisted the help of one of the world-leading NMR QC groups a couple of years ago to run a small molecular simulation algorithm, which failed because of the complexity of the pulse sequence required (including refocusing, etc.). Now if Scott wanted to solve a Maximum Independent Set problem on Orion, here is what he has to do: (1) load a graph into a database program; (2) type Find MIS in graph; (3) voila the output arrives on his terminal. He can do that from his couch at home, as long as he has an internet connection.

So if by practical I mean “concerned with actual use or practice” , which I believe is an accepted definition of the word, Orion is practical (a non-expert can use it) whereas all other QCs are not (only experts have a chance of being able to use them). Ergo the world’s first practical QC.

If anybody thinks this is an unreasonable argument I’d like to hear why–seems good to me!

Here’s another quote from the article:

Umesh Vazirani, a professor of computer science at the University of California, Berkeley, said, “D-Wave is misleading the public by calling their device ‘a practical quantum computer.’ The whole point of quantum computing is achieving a large speedup over classical computers, something that D-Wave hasn’t accomplished.”

Re. the first point, either Umesh is wrong or my argument presented in the prior segment is wrong. Re. the second: This is obviously the objective of building real QCs. On this point I agree. (although I would mention as an aside that there are many other “points” to QC, and this may not even be a majority opinion). However it seems pretty clear to me that on the path to achieving this objective, an effort will create machines that are–under the actual accepted definition of what a QC is (see eg. wikipedia)–real QCs that aren’t sufficiently powerful to have reached this end goal. Like Orion.

Describing what we are doing along the way has clear benefits in two related dimensions: (1) it helps us find matches with potential users, in order to focus development on high-value applications; (2) it allows the project to be properly financed. As Umesh knows from his experience in start-ups and industry, the main failure mechanism for any start-up is under capitalization.

More on this later!